CN109828182A - A kind of network system accident analysis method for early warning based on failure modes processing - Google Patents
A kind of network system accident analysis method for early warning based on failure modes processing Download PDFInfo
- Publication number
- CN109828182A CN109828182A CN201811446540.7A CN201811446540A CN109828182A CN 109828182 A CN109828182 A CN 109828182A CN 201811446540 A CN201811446540 A CN 201811446540A CN 109828182 A CN109828182 A CN 109828182A
- Authority
- CN
- China
- Prior art keywords
- failure
- early warning
- time series
- model
- series data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention discloses a kind of network system accident analysis method for early warning based on failure modes processing, it is related to electric power network technique field, the following steps are included: data acquisition combing step: obtaining all kinds of fault datas of network system in certain time period, and fault data is constituted time series data;Fault type classifying step: judging whether the generation of failure has periodicity according to time series data, periodically classifies to fault type according to whether failure has;Warning step: prediction model is established respectively to sorted failure, different types of time series data is inputted in corresponding prediction model, obtain prediction result, centered on prediction result, early warning red line is established by width of the standard deviation of presupposition multiple, is that foundation judges whether to early warning with early warning red line.Different prediction models is established, for different types of failure different types of failure is prejudged and be warned.
Description
Technical field
The present invention relates to electric power network technique field, in particular to a kind of network system accident analysis based on failure modes processing
Method for early warning.
Background technique
In recent years, with computer and the communication technology continuous development and apply to administration of power networks, substantially increase power grid
The efficiency of management of system has ensured network system safe and stable operation.In grid collapses or when being disturbed, digital guarantor
The intelligent electronic devices such as shield and fault oscillograph will record a large amount of data.Such as when grid power transmission route breaks down, line
Relay protection device, the oscillograph at road both ends generate fault message, are summarized by above-mentioned relay protection fault information and are analyzed, electricity
The methods of net breakdown judge, failure information system can form the fault message at the secondary faulty line both ends.
It is horizontal to reach higher electric service, shorten the breakdown repair time as far as possible, after power supply company needs look-ahead
Continue several days number of faults, to shift to an earlier date config failure repairing resource.Therefore, realize that distribution network failure quantity is accurate
Short-term forecast, it is horizontal to electric service is improved, the repairing level of resources utilization is promoted, is of great significance.
When network system scale more voluminous, each operation system of control centre is transferred to from each substation and all kinds of is set
Standby event, alarm, failure and the data logging generated is magnanimity, and a large amount of manpower intervention is needed to be handled.This just leads
It has caused positioning and the predicted time of failure longer, has affected the localization of fault time, so as to cause event of failure extension, problem is not
It can solve to prevent in advance in time, user experience is bad.In order to solve this problem, it needs a set of accuracy high and adapts to
Property strong intelligent trouble method for early warning, these a large amount of data can be utilized, adaptability prejudge different types of failure
With warning.
Summary of the invention
The invention is intended to provide a kind of network system accident analysis method for early warning based on failure modes processing, for difference
The failure of type establishes different prediction models, different types of failure is prejudged and be warned.
In order to solve the above technical problems, base case provided by the invention is as follows:
A kind of network system accident analysis method for early warning based on failure modes processing, comprising the following steps:
Data acquisition combing step: all kinds of fault datas of network system in certain time period are obtained, and by fault data
Constitute time series data;
Fault type classifying step: judging whether the generation of failure has periodicity according to time series data, according to event
Whether barrier, which has, is periodically classified to fault type;
Warning step: prediction model is established to sorted failure respectively, different types of time series data is inputted
In corresponding prediction model, prediction result is obtained, centered on prediction result, is established using the standard deviation of presupposition multiple as width pre-
Alert red line is that foundation judges whether to early warning with early warning red line.
Technical solution of the present invention obtains all kinds of fault datas of network system in certain time period, and by fault data
Time series data, how many class fault data, with regard to how many time series data constituted;Judged according to time series data
Whether the generation of failure has periodicity, and whether have with failure is periodically that standard classifies to fault type;And respectively
Different prediction models is established to different types of failure, and different failures is predicted using different prediction models, with
Centered on prediction result, early warning red line is established by width of the standard deviation of presupposition multiple, is that foundation judges whether with early warning red line
Early warning is carried out, by establishing early warning red line, relevant staff can intuitively observe the number of faults of each moment network system
Whether amount is more than early warning red line, and when number of faults is more than early warning red line, staff can be according to different failure and warning
Reason generates related counte-rplan, thus the generation of effectively trouble saving.In contrast to traditional single algorithm model prediction mode,
The present invention establishes different prediction models for different types of failure, substantially increases the adaptability and accuracy of algorithm.?
It is more accurate to more efficient in following time series forecasting, improve the reliability and practicability of fault pre-alarming.
Further, the fault type classifying step further include: if failure has periodically, for routinely failure;If
Failure does not have periodically, then is importance failure;
The warning step specifically includes:
Fault type judgment step: fault type is judged for routinely failure or importance failure, if routinely event
Barrier, then execute S101 and S102;If importance failure, then S103 is executed;
S101: acquisition time sequence data establishes the first prediction model, and time series data is inputted the first prediction model
In, constantly model parameter is optimized by evaluation index, and obtain the optimal model parameters of prediction model;
S102: prediction result is obtained according to optimal model parameters, centered on prediction result, with the standard deviation of presupposition multiple
Early warning red line is established for width, is that foundation judges whether to early warning with early warning red line;
S103: acquisition time sequence data establishes the second prediction model, and time series data is inputted the second prediction model
In, it obtains prediction result and early warning red line is established by width of the standard deviation of presupposition multiple, centered on prediction result with early warning
Red line is that foundation judges whether to early warning.
Fault type judgment step judges fault type for routinely failure or importance failure, and routinely failure refers to
Frequent occurrence and have periodically, importance failure, which refers to, seldom to be occurred and do not have periodically.
When for routinely failure, the first prediction model is established, time series data is inputted in the first prediction model, is led to
It crosses evaluation index constantly to optimize model parameter, and obtains the optimal model parameters of prediction model;Joined according to optimal models
Number obtains prediction result and establishes early warning red line by width of the standard deviation of presupposition multiple centered on prediction result, red with early warning
Line is that foundation judges whether to early warning.First prediction model constantly optimizes model parameter by evaluation index, also
It is to review one's lessons by oneself formula iterative learning correction model using failure, improves the accuracy of model.
When for the property wanted failure, the second prediction model is established, time series data is inputted in the second prediction model, is obtained
Prediction result establishes early warning red line by width of the standard deviation of presupposition multiple centered on prediction result, with early warning red line be according to
It is judged that whether carrying out early warning.It does not need namely to optimize model parameter.
Further, in the S101: establishing the first prediction model using SARIMA algorithm;In the S103: being calculated using MA
Method establishes the second prediction model.
When time series table reveals seasonal variety and linear trend, random seaconal model and model can be combined into season
Time series models, that is, model is saved to describe the time series, referred to as SARIMA.SARIMA model is a kind of short-time forecasting model,
Core element is the processing to data, while using the error for going value to generate after being fitted as Essential Elements Of Analysis, advantage outstanding is
The precision of short-term prediction result is higher.For the particularity of importance failure, mould is established using the higher MA algorithm of fitting degree
Type, compared to SARIMA model, the foundation of MA model is more convenient.
Further, in the S1: constantly being optimized to model parameter by AIC evaluation index.
AIC information criterion, that is, Akaike information criterion is measure statistical models fitting Optimality
A kind of standard of (Goodness of fit).
Further, the fault type judgment step: if routinely failure, then S101, S102 and S104 are executed;
S104: acquisition time sequence data samples time series data in the predetermined time in each period, and counts
The slope value for obtaining the time series data at two neighboring moment is calculated, and G-bar is calculated according to the number of predetermined time
Value carries out early warning when mean slope values are greater than preset slope threshold value.
For the time series periodically amplified, prediction result be also periodically amplify, if execute S101 and
S102 step, obtained early warning red line can also follow prediction result periodically to amplify, and being possible in this way can be beyond the normal of prediction
Range, so need another judgment mode, i.e. execution S104, each period predetermined time to time series data into
Row sampling, and the slope value of the time series data at two neighboring moment is calculated, and calculate according to the number of predetermined time
It obtains mean slope values, when mean slope values are greater than preset slope threshold value, illustrates that growth rate is too fast, carry out early warning.
Further, the S101 is specifically included:
S101-1: acquisition time sequence data draws to time series data, judges whether the time series data is flat
Steady time series then executes S101-2 if nonstationary time series;If stationary time series, then S101-3 is executed;
S101-2: d order difference operation to be carried out first to nonstationary time series, turn to stationary time series, then execute
S101-3;
S101-3: constantly model parameter is optimized for evaluation index so that AIC is optimal, and obtains prediction model most
Excellent model parameter.
Only time series data is stationary time series, can just bring the first prediction model into and be calculated.
Detailed description of the invention
Fig. 1 is a kind of process of the network system accident analysis method for early warning embodiment based on failure modes processing of the present invention
Figure.
Specific embodiment
It is further described below by specific embodiment:
Embodiment one
As shown in Figure 1, a kind of network system accident analysis method for early warning based on failure modes processing of the present invention, including with
Lower step:
Data acquisition combing step: all kinds of fault datas of network system in certain time period are obtained, and by fault data
Constitute time series data;Specifically, multiple monitoring nodes are arranged in distribution on power grid transmission line road, for example, monitoring node is
The oscillograph at route both ends, what the present embodiment obtained is all kinds of number of faults of each monitoring node of network system in certain time period
According to;How many class fault data, with regard to how many time series data, for example, there is the appearance of 3 class error codes, then representing has 3 kinds of events
Barrier, then have 3 time serieses, and specifically, fault data is the number of error code;The present embodiment obtain be within one day for 24 hours
Network system all kinds of failures time series data, prediction is following 24 hours number of faults;
Fault type classifying step: judging whether the generation of failure has periodicity according to time series data, according to event
Whether barrier, which has, is periodically classified to fault type;If failure has periodically, for routinely failure;If failure does not have
There is periodicity, is then importance failure;
Warning step: prediction model is established to sorted failure respectively, different types of time series data is inputted
In corresponding prediction model, prediction result is obtained, centered on prediction result, is established using the standard deviation of presupposition multiple as width pre-
Alert red line is that foundation judges whether to early warning with early warning red line.
In the present embodiment, warning step is specifically included:
Fault type judgment step: fault type is judged for routinely failure or importance failure, if routinely event
Barrier, then execute S101, S102 and S104;If importance failure, then S103 is executed;
S101: acquisition time sequence data establishes the first prediction model using SARIMA algorithm, and time series data is defeated
Enter in the first prediction model, constantly model parameter is optimized by AIC evaluation index, and obtains the optimal mould of prediction model
Shape parameter;
S102: prediction result is obtained according to optimal model parameters, centered on prediction result, with the standard deviation of presupposition multiple
Early warning red line is established for width, is that foundation judges whether to early warning with early warning red line;The present embodiment presupposition multiple is 2 times;
S103: acquisition time sequence data establishes the second prediction model using MA algorithm, by time series data input the
In two prediction models, obtains prediction result and it is red to establish early warning using the standard deviation of presupposition multiple as width centered on prediction result
Line is that foundation judges whether to early warning with early warning red line;
S104: acquisition time sequence data samples time series data in the predetermined time in each period, and counts
The slope value for obtaining the time series data at two neighboring moment is calculated, and G-bar is calculated according to the number of predetermined time
Value carries out early warning when mean slope values are greater than preset slope threshold value.
One, for routinely failure
Routinely failure is divided into periodically amplification failure and aperiodicity amplification failure.
A, for the failure of aperiodicity amplification:
1, ARIMA algorithm prediction model is established
The full name of ARIMA model is called ARMA model, is denoted as ARIMA (p, d, q).Its meaning are as follows: assuming that
One random process contains d unit root, and a stable auto regressive moving average mistake can be transformed to after d difference
Journey, then the random process is known as single product (whole) autoregressive moving-average (ARMA) process.General type is
Φ(L)Δdxt=δ+Θ (L) ut
Wherein, xtFor former sequence, L indicates backward shift operator, Δd=(1-L)dFor d order difference, Φ (L)=1- Φ1L-Φ2L2-…-ΦpLp, Θ (L)=1- θ1L-θ2L2-…θpLp, utFor zero-mean white noise series.
2, SARIMA model (the first prediction model) is established
When time series table reveals seasonal variety and linear trend, random seaconal model and model can be combined into season
Time series models, that is, model is saved to describe the time series, referred to as SARIMA.SARIMA model is a kind of short-time forecasting model,
Core element is the processing to data, while using the error for going value to generate after being fitted as Essential Elements Of Analysis, advantage outstanding is
The precision of short-term prediction result is higher.The general type of SARIMA model is expressed as
Φp(L)AP(LT)(ΔdΔTxt)=Θ (L) BQ(LT)ut
T indicates the period of change of seasonal sequence in formula;L indicates lag operator;Φp(L)、AP(LT) respectively indicate non-season
Section and season autoregression multinomial;Θ(L),BQ(LT) then respectively indicate non-season and season rolling average multinomial;Subscript P, Q,
P, q respectively indicates the maximum lag order in season and non-season autoregression, moving average operator;D, D respectively indicate non-season and
Seasonal difference number in practical applications, if former sequence includes simultaneously tendency and seasonality, is represented by
SeasonalARIMA (p, d, q) × (P, D, Q, T) model.
Algorithm key step is as follows:
A, sequence data x is obtainedt, according to the scatter plot of time series, auto-correlation function and partial autocorrelation function figure with ADF
Unit root test its variance, trend and its Rules of Seasonal Changes, identify the stationarity of sequence.Pass through difference and season
Former sequence is converted into a stable sequence by difference.
D order difference operation will be carried out first for nonstationary time series, turn to stationary time series.
In formula, wtFor stationary sequence.
B, w is obtainedt~ARMA (p, q), model form are
C, using AIC as evaluation index, continuous iteration changes SARIMA model parameter, and acquisition keeps AIC index optimal
SARIMA model parameter obtains model xt~SARIMA (p, d, q) × (P, D, Q, T), wherein T is the fixed cycle.
3, the model optimization based on AIC evaluation index
AIC information criterion, that is, Akaike information criterion is measure statistical models fitting Optimality
A kind of standard of (Goodness of fit).Its calculation formula is as follows:
WhereinFor penalty factor.
Iteration is to repeat the activity of feedback procedure, and purpose is typically to approaching required target or result.We refer to R
Arima standard in language, with (p, d, q)=(5,2,5), (P, D, Q)=(5,2,5) are maximum value, and 1 is step-length, is successively subtracted
Parameters value in small (p, d, q) and (P, D, Q) realizes p*d*q*P*D*Q (i.e. 5*2*5*5*2*5=2500) secondary traversal meter
It calculates, then choosing in the result makes the smallest SARIMA optimal model parameters of AIC index.
B, for the failure periodically amplified:
For the time series periodically amplified, prediction result be also periodically amplify, if execute S101 and
S102 step, obtained early warning red line can also follow prediction result periodically to amplify, and being possible in this way can be beyond the normal of prediction
Range, so needing another judgment mode, that is, the mode for the time series slope value for taking calculating cycle to amplify exists
The predetermined time in each period samples, and the slope value of the time series data at two neighboring moment is calculated, and according to pre-
If mean slope values are calculated in the number at moment, when mean slope values are greater than preset slope threshold value, illustrate growth rate
It is too fast, carry out early warning.
Such as branch mailbox was carried out for 24 hours to one day, it is divided into 24 casees, predetermined time is 24 integral points, and each integral point is to error code
Number is sampled, and carries out simple regression using the error code number at corresponding moment, obtains 24 trend to get to 24 slopes
Value, the mean slope values for the integral cycle amplification time sequence among as one day of averagely getting off.
To sum up, as long as routinely failure, S101, S102 and S104 will be passed through, when actual time series is beyond pre-
When the range or mean slope values of alert red line are greater than preset slope threshold value, early warning can be all carried out.
Two, it is directed to importance failure
1, MA algorithm model (the second prediction model) is established
MA (q) model also known as q rank moving average model(MA model), model expression are as follows
xt=μ+ut+θ1ut-1+θ2ut-2+…+θqut-q
xt- μ=(1+ θ1L+θ2L2+…+θqLq)ut=Θ (L) ut
Wherein utIt is white-noise process.
Compared to SARIMA model, the foundation of MA model is more convenient.Firstly, obtaining sequence data xt, to map data,
It sees whether as stationary sequence.Since a large amount of priori knowledges show that importance failure is commonly stationary sequence, it is used directly for
Auto-correlation coefficient is sought, without carrying out d order difference operation.
Seek MA (q) auto-correlation coefficient ρk:
xt=μ+ut+θ1ut-1+θ2ut-2+…+θqut-q
As k > q, ρk=0, xtWith xt+kUncorrelated, this phenomenon is known as truncation, thus can according to auto-correlation coefficient whether
It is always the 0 order q to judge MA (q) model since certain point, it is determined that after order q, prediction result can be obtained.
For example, to MA (1) process Xt=εt-θεt-1, the auto-correlation function that can find out MA (1) process is
As it can be seen that as k > 1, ρk> 0, i.e. xtWith xt+kUncorrelated, MA (1) auto-correlation function is truncation.
Three, the foundation of early warning red line
It is width with 2 times of standard deviations centered on prediction result for the failure and importance failure of aperiodicity amplification
Degree establishes early warning red line, is that foundation judges whether to early warning with early warning red line;In actual electric network operation, when actual failure
Time series has exceeded early warning red line or the mean slope values of actual fault time sequence have exceeded preset slope threshold
Value, i.e., warn various failures, and staff can count mistake, staff can according to different failure and
Reason is warned to generate related counte-rplan, thus the generation of effectively trouble saving.It is predicted in contrast to traditional single algorithm model
Mode, the present invention establish different prediction models for different types of failure, substantially increase the adaptability of algorithm and accurate
Property.It is more efficient in following time series forecasting, it is more accurate, improve the reliability and practicability of fault pre-alarming.
Embodiment two
The difference between this embodiment and the first embodiment lies in, further includes:
Database is previously stored with the address information and monitoring node line information detected of each monitoring node,
Line information includes set-up time and address properties, and address properties include indoor and outdoor, the address information of the monitoring node
It is corresponded with line information;Monitoring node in the present embodiment is the oscillograph at route both ends;Specifically, line information are as follows:
Monitoring node A detection is to number the route for being 0012, and the set-up time of the route is on April 12nd, 2018, which is located at
It is indoor;Monitoring node B detection is to number the route for being 0045, and the set-up time of the route is on May 1st, 2018, the route
Positioned at open air;Monitoring node C detection is to number the route for being 0036, and the set-up time of the route is on April 12nd, 2018, should
Route is located at open air;
Early warning judgment step: judge whether actual fault time sequence is more than early warning red line, and judge G-bar
Whether value is greater than slope threshold value;The early warning judges to watch as staff's naked eyes, such as what is presented on display screen is early warning
Red line and actual fault time sequence are pressed when observing that actual fault time sequence is more than early warning red line by starting
Button starts next step step;The present embodiment judges automatically whether actual fault time sequence is more than that early warning is red using computer
Line, and judge whether mean slope values are greater than slope threshold;
Route transfers step: when actual fault time sequence is more than that early warning red line or mean slope values are greater than slope threshold
When value, according to one-to-one line information, and root in the address information matching database for the monitoring node for sending fault data
All line informations identical with the route set-up time are transferred from database according to line information;When actual fault time sequence
When column are greater than slope threshold value more than early warning red line or mean slope values, that is, there is exception in failure, since database is pre-
It is first stored with the address information of each monitoring point, when a certain monitoring node breaks down, system has just obtained the monitoring node
Address information find the monitoring node institute then according to one-to-one line information in the address information matching database
The route of detection, and the All other routes installed with the route with batch are found, that is, the All other routes installed on the same day;Example
Such as: assuming that break down is monitoring node A, being deployed into the number that the route for being 0012 with number is installed on the same day at this time is
0036 route;
Address properties judgment step: the address properties in the line information transferred are judged to be indoor or outdoor, if address
Attribute be it is indoor, then execute S201;If address properties are open air, S202 is executed;The namely route that it is 0036 that judgement, which is numbered,
Address properties;
S201: according to the line information transferred by line marker be level-one easy break-down route;
S202: according to the line information transferred by line marker be three-level easy break-down route;The address of 0036 route
Attribute is open air, so being three-level easy break-down route by the line marker, series is higher, and the probability of easy break-down is bigger, because
Service life for the route of same batch installation is identical, predicts that route is possible to break down by the set-up time, from
And pay close attention to the route;Route will receive outdoors to expose to the sun and rain, so for indoor route, failure
Probability wants higher, so series wants higher;
Address range partiting step: it using the midpoint of three-level easy break-down route as the center of circle, is drawn by radius of pre-determined distance value
Circle from other all monitoring nodes transferred in drawn circle range in database, and judges that the monitoring node transferred is detected
Whether the set-up time of route is prior to the monitoring node of the failure route set-up time detected;If prior to failure
The monitoring node route set-up time detected, then execute S203;If the monitoring node for being later than failure route detected
Set-up time then executes S204;On the basis of the set-up time predict failure on the basis of, this step be using environmental factor come
Predict a possibility that failure occurs, the environment as suffered by the route in same section is identical, so route goes wrong
A possibility that it is roughly the same;Specifically, it using the midpoint of three-level easy break-down route as the center of circle, is drawn by radius of pre-determined distance value
Circle is circle from other all monitoring nodes transferred in drawn circle range in database, that is, with the midpoint of 0036 route
The heart, it is assumed that the monitoring node found within the limits prescribed has D, E, and monitoring node D detection is to number the route for being 0078, should
The set-up time of route is on January 1st, 2018;Monitoring node D detection is route that number is 0026, when the installation of the route
Between be on September 5th, 2018, then judge number be 0078 and 0026 route set-up time whether prior to number be 0012
The set-up time of route;
S203: being level Four easy break-down route by corresponding line marker;Due to the route set-up time that number is 0078
The route for being 0012 prior to number, so the route that number is 0078 is marked as level Four easy break-down route;
S204: being second level easy break-down route by corresponding line marker;Due to the route set-up time that number is 0026
It is later than the route that number is 0012, so the route that number is 0026 is marked as second level easy break-down route, because when installation
Between if route prior to failure, illustrate the also longer using the time than the route of failure using the time of the route,
A possibility that failure, is very big, needs to pay close attention to;And the set-up time if the route for being later than failure, that is,
It is later than the set-up time of three-level easy break-down route, so being marked as second level easy break-down route.
On the basis of predicting failure on the basis of the set-up time, environmental factor is recycled to predict the possibility of failure generation
Property, to divide the grade of easy break-down to the route of different set-up times and area, predict that failure is sent out according to grade difference
The height of a possibility that raw, staff can take in advance different arrange to different routes according to different fault levels
It applies, improves the accuracy and reliability of failure predication.
What has been described above is only an embodiment of the present invention, and the common sense such as well known specific structure and characteristic are not made herein in scheme
Excessive description.It, without departing from the structure of the invention, can be with it should be pointed out that for those skilled in the art
Several modifications and improvements are made, these also should be considered as protection scope of the present invention, these all will not influence what the present invention was implemented
Effect and patent practicability.The scope of protection required by this application should be based on the content of the claims, in specification
The records such as specific embodiment can be used for explaining the content of claim.
Claims (6)
1. a kind of network system accident analysis method for early warning based on failure modes processing, which comprises the following steps:
Data acquisition combing step: all kinds of fault datas of network system in certain time period are obtained, and fault data is constituted
Time series data;
Fault type classifying step: judge whether the generation of failure has periodicity according to time series data, be according to failure
It is no to classify with periodical to fault type;
Warning step: establishing prediction model to sorted failure respectively, and different types of time series data is inputted and is corresponded to
Prediction model in, obtain prediction result it is red to establish early warning using the standard deviation of presupposition multiple as width centered on prediction result
Line is that foundation judges whether to early warning with early warning red line.
2. the network system accident analysis method for early warning according to claim 1 based on failure modes processing, feature exist
In the fault type classifying step further include: if failure has periodically, for routinely failure;If failure does not have week
Phase property is then importance failure;
The warning step specifically includes:
Fault type judgment step: judge fault type for routinely failure or importance failure, if routinely failure, then
Execute S101 and S102;If importance failure, then S103 is executed;
S101: acquisition time sequence data establishes the first prediction model, and time series data is inputted in the first prediction model,
Constantly model parameter is optimized by evaluation index, and obtains the optimal model parameters of prediction model;
S102: obtaining prediction result according to optimal model parameters, is width with the standard deviation of presupposition multiple centered on prediction result
Degree establishes early warning red line, is that foundation judges whether to early warning with early warning red line;
S103: acquisition time sequence data establishes the second prediction model, and time series data is inputted in the second prediction model,
It obtains prediction result and early warning red line is established by width of the standard deviation of presupposition multiple, centered on prediction result with early warning red line
Early warning is judged whether to for foundation.
3. the network system accident analysis method for early warning according to claim 2 based on failure modes processing, feature exist
In in the S101: establishing the first prediction model using SARIMA algorithm;In the S103: it is pre- to establish second using MA algorithm
Survey model.
4. the network system accident analysis method for early warning according to claim 2 based on failure modes processing, feature exist
In in the S101: constantly being optimized to model parameter by AIC evaluation index.
5. the network system accident analysis method for early warning according to claim 2 based on failure modes processing, feature exist
In the fault type judgment step: if routinely failure, then executing S101, S102 and S104;
S104: acquisition time sequence data samples time series data in the predetermined time in each period, and calculates
The slope value of the time series data at two neighboring moment out, and mean slope values are calculated according to the number of predetermined time,
When mean slope values are greater than preset slope threshold value, early warning is carried out.
6. the network system accident analysis method for early warning according to claim 4 based on failure modes processing, feature exist
In the S101 is specifically included:
S101-1: acquisition time sequence data draws to time series data, when judging whether the time series data is steady
Between sequence if nonstationary time series then execute S101-2;If stationary time series, then S101-3 is executed;
S101-2: d order difference operation to be carried out first to nonstationary time series, turn to stationary time series, then execute S101-3;
S101-3: constantly model parameter is optimized for evaluation index so that AIC is optimal, and obtains the optimal mould of prediction model
Shape parameter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811446540.7A CN109828182B (en) | 2018-11-29 | 2018-11-29 | Power grid system fault analysis and early warning method based on fault classification processing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811446540.7A CN109828182B (en) | 2018-11-29 | 2018-11-29 | Power grid system fault analysis and early warning method based on fault classification processing |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109828182A true CN109828182A (en) | 2019-05-31 |
CN109828182B CN109828182B (en) | 2021-01-19 |
Family
ID=66859192
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811446540.7A Active CN109828182B (en) | 2018-11-29 | 2018-11-29 | Power grid system fault analysis and early warning method based on fault classification processing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109828182B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110362267A (en) * | 2019-07-11 | 2019-10-22 | 深圳市科航科技发展有限公司 | A kind of screening machine inputting equipment system |
CN110672332A (en) * | 2019-09-10 | 2020-01-10 | 上海电力大学 | Gas turbine fault early warning system based on SARIMA model |
CN111314110A (en) * | 2020-01-17 | 2020-06-19 | 南京大学 | Fault early warning method for distributed system |
CN112115180A (en) * | 2020-09-11 | 2020-12-22 | 国网山东省电力公司枣庄供电公司 | Power grid accident prediction method based on big data |
CN112215108A (en) * | 2020-09-29 | 2021-01-12 | 三一专用汽车有限责任公司 | Mixer truck fault prejudging method and device and computer readable storage medium |
CN112232381A (en) * | 2020-09-25 | 2021-01-15 | 国网上海市电力公司 | Model parameter post-processing method and device for leading load parameter noise identification |
CN112465237A (en) * | 2020-12-02 | 2021-03-09 | 浙江正泰电器股份有限公司 | Fault prediction method, device, equipment and storage medium based on big data analysis |
CN114414938A (en) * | 2021-12-22 | 2022-04-29 | 南通联拓信息科技有限公司 | Dynamic response method and system for power distribution network fault |
CN114444739A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司 | Digital smart power grid region management system and method |
CN114697203A (en) * | 2022-03-31 | 2022-07-01 | 浙江省通信产业服务有限公司 | Network fault pre-judging method and device, electronic equipment and storage medium |
CN117332857A (en) * | 2023-09-19 | 2024-01-02 | 上海聚数信息科技有限公司 | Multi-source data-based power grid data automatic management system and method |
CN117491810A (en) * | 2023-12-27 | 2024-02-02 | 国网山东省电力公司济宁供电公司 | Overvoltage flexible inhibition data acquisition method and system |
CN117609740A (en) * | 2024-01-23 | 2024-02-27 | 青岛创新奇智科技集团股份有限公司 | Intelligent prediction maintenance system based on industrial large model |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103116531A (en) * | 2013-01-25 | 2013-05-22 | 浪潮(北京)电子信息产业有限公司 | Storage system failure predicting method and storage system failure predicting device |
CN105718958A (en) * | 2016-01-27 | 2016-06-29 | 江苏省电力公司检修分公司 | Breaker fault diagnosis method based on linear discriminant analysis and support vector machine |
JP2017049093A (en) * | 2015-09-01 | 2017-03-09 | 東日本旅客鉄道株式会社 | Fault point orientation system and method for dc electric railway feeding circuit |
CN108320043A (en) * | 2017-12-19 | 2018-07-24 | 江苏瑞中数据股份有限公司 | A kind of distribution network equipment state diagnosis prediction method based on electric power big data |
CN108830745A (en) * | 2018-06-29 | 2018-11-16 | 国网上海市电力公司 | Power grid cascading fault diagnosis, early warning, assessment system based on monitoring information |
-
2018
- 2018-11-29 CN CN201811446540.7A patent/CN109828182B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103116531A (en) * | 2013-01-25 | 2013-05-22 | 浪潮(北京)电子信息产业有限公司 | Storage system failure predicting method and storage system failure predicting device |
JP2017049093A (en) * | 2015-09-01 | 2017-03-09 | 東日本旅客鉄道株式会社 | Fault point orientation system and method for dc electric railway feeding circuit |
CN105718958A (en) * | 2016-01-27 | 2016-06-29 | 江苏省电力公司检修分公司 | Breaker fault diagnosis method based on linear discriminant analysis and support vector machine |
CN108320043A (en) * | 2017-12-19 | 2018-07-24 | 江苏瑞中数据股份有限公司 | A kind of distribution network equipment state diagnosis prediction method based on electric power big data |
CN108830745A (en) * | 2018-06-29 | 2018-11-16 | 国网上海市电力公司 | Power grid cascading fault diagnosis, early warning, assessment system based on monitoring information |
Non-Patent Citations (1)
Title |
---|
卢世成: "基于数据流的电网设备风险识别与预警模型研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110362267A (en) * | 2019-07-11 | 2019-10-22 | 深圳市科航科技发展有限公司 | A kind of screening machine inputting equipment system |
CN110362267B (en) * | 2019-07-11 | 2020-11-10 | 深圳市科航科技发展有限公司 | Input equipment system of security check machine |
CN110672332A (en) * | 2019-09-10 | 2020-01-10 | 上海电力大学 | Gas turbine fault early warning system based on SARIMA model |
CN111314110A (en) * | 2020-01-17 | 2020-06-19 | 南京大学 | Fault early warning method for distributed system |
CN112115180A (en) * | 2020-09-11 | 2020-12-22 | 国网山东省电力公司枣庄供电公司 | Power grid accident prediction method based on big data |
CN112232381A (en) * | 2020-09-25 | 2021-01-15 | 国网上海市电力公司 | Model parameter post-processing method and device for leading load parameter noise identification |
CN112232381B (en) * | 2020-09-25 | 2024-03-01 | 国网上海市电力公司 | Model parameter post-processing method and device for dominant load parameter noise identification |
CN112215108A (en) * | 2020-09-29 | 2021-01-12 | 三一专用汽车有限责任公司 | Mixer truck fault prejudging method and device and computer readable storage medium |
CN112465237B (en) * | 2020-12-02 | 2023-04-07 | 浙江正泰电器股份有限公司 | Fault prediction method, device, equipment and storage medium based on big data analysis |
CN112465237A (en) * | 2020-12-02 | 2021-03-09 | 浙江正泰电器股份有限公司 | Fault prediction method, device, equipment and storage medium based on big data analysis |
CN114414938A (en) * | 2021-12-22 | 2022-04-29 | 南通联拓信息科技有限公司 | Dynamic response method and system for power distribution network fault |
CN114697203A (en) * | 2022-03-31 | 2022-07-01 | 浙江省通信产业服务有限公司 | Network fault pre-judging method and device, electronic equipment and storage medium |
CN114697203B (en) * | 2022-03-31 | 2023-07-25 | 浙江省通信产业服务有限公司 | Network fault pre-judging method and device, electronic equipment and storage medium |
CN114444739A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司 | Digital smart power grid region management system and method |
CN114444739B (en) * | 2022-04-11 | 2022-07-29 | 广东电网有限责任公司 | Digital smart power grid region management system and method |
CN117332857A (en) * | 2023-09-19 | 2024-01-02 | 上海聚数信息科技有限公司 | Multi-source data-based power grid data automatic management system and method |
CN117332857B (en) * | 2023-09-19 | 2024-04-02 | 上海聚数信息科技有限公司 | Multi-source data-based power grid data automatic management system and method |
CN117491810A (en) * | 2023-12-27 | 2024-02-02 | 国网山东省电力公司济宁供电公司 | Overvoltage flexible inhibition data acquisition method and system |
CN117609740A (en) * | 2024-01-23 | 2024-02-27 | 青岛创新奇智科技集团股份有限公司 | Intelligent prediction maintenance system based on industrial large model |
Also Published As
Publication number | Publication date |
---|---|
CN109828182B (en) | 2021-01-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109828182A (en) | A kind of network system accident analysis method for early warning based on failure modes processing | |
CN109146093B (en) | Power equipment field investigation method based on learning | |
CN108765187B (en) | Power failure fault auxiliary study and judgment method, device and system | |
EP3913555A1 (en) | Power distribution network operation aided decision-making analysis system and method | |
US9158035B2 (en) | System and method of automated acquisition, correlation and display of power distribution grid operational parameters and weather events | |
CN112036610A (en) | Intelligent cable early warning platform and early warning method based on big data analysis | |
US10809885B2 (en) | Identifying and presenting related electrical power distribution system events | |
CN106446016A (en) | Parallel association rule mining-based method for predicting running reliability of power distribution network | |
CN113162232B (en) | Power transmission line equipment risk assessment and defense decision system and method | |
CN112136161A (en) | System and method for intelligent alarm grouping | |
US11416119B2 (en) | Locating a power line event downstream from a power line branch point | |
CN106373032B (en) | The high-incidence region discrimination method of Distribution Network Failure based on big data | |
CN105488740A (en) | Method and system for evaluating risk grade of operation mode of urban power distribution network | |
CN116451876B (en) | Power distribution network fault prediction and active overhaul system based on artificial intelligence | |
CN112085618A (en) | Frequent item set-based power distribution equipment fault factor relevance mining method | |
US10852341B2 (en) | Composite fault mapping | |
US10837995B2 (en) | Composite fault mapping | |
Al-Kanj et al. | A probability model for grid faults using incomplete information | |
CN113344735B (en) | Disaster prevention and reduction monitoring and early warning system of power grid equipment | |
CN107294205B (en) | Substation state monitoring method based on information protection master station data | |
CN113327174A (en) | Distribution transformer load prediction method | |
CN112016739A (en) | Fault detection method and device, electronic equipment and storage medium | |
US11158007B2 (en) | Dynamic energy consumption and harvesting with feedback | |
Kabir et al. | Power outage prediction using data streams: An adaptive ensemble learning approach with a feature‐and performance‐based weighting mechanism | |
CN111313355A (en) | Method for updating monitoring signal event rule under manual supervision |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |